Research On Wind Farm Output Forecasting Using Dynamic Neural Networks And Application

Wind power is intermittent, stochastic and uncontrollable. With more wind power being integrated into grid systems, forecasting wind power is playing a more significant role in power system operation, wind farm operation and electricity market. Short-term forecasting of wind power is useful to primary and secondly reserve control and unit commitment for power system operation, and is useful to wind turbine control, unit commitment and scheduling, maintenance scheduling and trade strategy for wind farm. Consequently, it is necessary to researching on wind power forecasting and its application.Based on numerical weather predictions, wind power forecasting models are built using neural networks. In order to improve neural networks'trainning performance, elements affecting wind power forecasting are discussed and choosed in detail and neural networks are coupled with spatial correlation approach. On account of neural saturated problem, original data, such as wind speed and wind direction, are normalized within the range [-1,1]. Elements affecting wind power forecasting are divided into several groups for building models.In order to improve the simulation of wind power with nolinear time series characteristic, two dynamic nueral networks are proposed, local recurrent time-delay network and global recurrent time-delay network respectively, to build 24 hours ahead wind power forecasting models. Orders of both dynamic neural networks are decided before training. Dynamic models are compared with two static models, BP and RBF nueral networks respectively. RMSE and MAE are used to value these models. Simulation results demonstrate that the dynamic nueral network wind power forecasting models outperform the static ones.The forecasting results are used to day-ahead unit commitment and scheduling condsider wind farm integration into grid systems. Dynamic programming priority approach is used for unit commitment, whose objective function is generating cost and reserve cost. Reserve is decided by the wind power forecsting precision, in order to discuss impacts of wind power forecasting on scheduling. Simulation results demonstrate that with high wind energy penetration, improving wind power forecasting precision can reduce reserve capacity and hence the cost of power systerm operation.